Every tier asks whether the tier below is good at its job. The manager tier's evaluation answers two questions at once, and we only noticed while trying to fix something else.
A manager's evaluation of another manager does two things: it sets that manager's score — the force they carry when evaluating trainers — and it directs energy, their influence over other managers. One evaluation, two consequences.
Those are different competencies. Backing accurate trainers is hands-on assessment of people doing the work. Backing accurate managers is judgment about judgment. Someone can be good at one and poor at the other, and nothing distinguishes them today.
The tier table shows the symptom. Every row reads "does this X back accurate Ys". The manager row reads "does this manager evaluate well" — well at what?
Why it's upstream of the energy work
We got here from negatives. In the energy layer a negative manager evaluation and no evaluation are identical — aura.py filters the flow to positive evaluations (lines 70 and 79), which is Adam's point on the call: no difference between setting someone to 0 and setting them to −4.
Every fix for that depends on which question energy is answering. If it's one question, negatives have to work inside a rating doing both jobs. If it's two, energy is answering the peer question — and that question has no negative answer, because it's an allocation of steering power rather than a judgment. You widen a pipe, narrow it, or close it; there's no negative amount of power to hand someone.
A sketch of the split, not a proposal
- A fifth row in the tier table — Steward — asking does this manager back accurate managers?
- Not a fourth evaluator tier. Managers stay the top tier, asked two questions; the three-tier rationale is untouched.
- Ratings unchanged everywhere: signed, 4× on negatives, unlimited, no cost to answering.
- Energy allocated rather than rated: no sign, defaults to none, dilutes as you spread it.
- And the reason it might be worth it — allocating energy badly has no consequence today. With a stewardship rating it does, because withholding energy from someone you know is a good steward is itself bad stewardship, and it becomes measurable.
The question
Adam — one question or two?
We kept all of this out of specs/how-aura-works.md, which states only what the code does and flags the ambiguity as open (§4, and open question 4). Either answer makes the follow-on work obvious; without it we'd be guessing.
Every tier asks whether the tier below is good at its job. The manager tier's evaluation answers two questions at once, and we only noticed while trying to fix something else.
A manager's evaluation of another manager does two things: it sets that manager's score — the force they carry when evaluating trainers — and it directs energy, their influence over other managers. One evaluation, two consequences.
Those are different competencies. Backing accurate trainers is hands-on assessment of people doing the work. Backing accurate managers is judgment about judgment. Someone can be good at one and poor at the other, and nothing distinguishes them today.
The tier table shows the symptom. Every row reads "does this X back accurate Ys". The manager row reads "does this manager evaluate well" — well at what?
Why it's upstream of the energy work
We got here from negatives. In the energy layer a negative manager evaluation and no evaluation are identical —
aura.pyfilters the flow to positive evaluations (lines 70 and 79), which is Adam's point on the call: no difference between setting someone to 0 and setting them to −4.Every fix for that depends on which question energy is answering. If it's one question, negatives have to work inside a rating doing both jobs. If it's two, energy is answering the peer question — and that question has no negative answer, because it's an allocation of steering power rather than a judgment. You widen a pipe, narrow it, or close it; there's no negative amount of power to hand someone.
A sketch of the split, not a proposal
The question
Adam — one question or two?
We kept all of this out of
specs/how-aura-works.md, which states only what the code does and flags the ambiguity as open (§4, and open question 4). Either answer makes the follow-on work obvious; without it we'd be guessing.